---
license: mit
task_categories:
- image-segmentation
- feature-extraction
- image-feature-extraction
language:
- en
tags:
- code
pretty_name: Visual Search Dataset
size_categories:
- 10K<n<100K
---
# Visual Search Image Dataset

## Overview

This dataset contains **10,343 real-world images organized into 139 semantic categories**.
Images represent animals, vehicles, people, food, nature scenes, technology, architecture, and everyday environments.

The dataset was created for experiments in **semantic image retrieval and multimodal embedding search** using models such as **CLIP** and vector search libraries like **FAISS**.

Images were automatically collected using the Python **icrawler** library from Bing Image Search and organized based on the search queries used during crawling.

---

## Dataset Statistics

* Total images: **10,343**
* Total categories: **139**
* Images per category: **~30–104**
* Average images per category: **~74**
* Image formats: **JPG, PNG, WEBP**
* Embedding ModelCLIP: **ViT-B/32**

---
## Categories
Animals (15 categories)
Dog running in park, cat sleeping on sofa, bird flying sky, horse galloping field, lion savanna, elephant wildlife, tiger jungle, dolphin ocean jumping, eagle hunting, wolf snow forest, bear fishing river, penguin antarctica, giraffe eating leaves, zebra herd, monkey tree climbing
Aviation (8 categories)
Airplane taking off runway, airplane landing airport, airplane flying clouds, fighter jet airshow, helicopter city, private jet luxury, airport terminal busy, airplane window view
Vehicles (12 categories)
Sports car highway night, motorcycle racing track, train mountain landscape, subway station crowd, truck highway, bicycle city street, sailboat ocean, speedboat lake, tractor farm field, ambulance city, police car street, fire truck emergency
People & Activities (22 categories)
Person riding bicycle street, chef cooking kitchen, musician playing guitar stage, athlete running marathon, dancer ballet performance, painter artist studio, surgeon operating room, teacher classroom, photographer nature, firefighter action, soldier military, astronaut space, diver underwater coral, climber mountain rock, surfer ocean wave, skier mountain slope, basketball player dunking, soccer player goal, tennis player court, swimmer pool race, boxer ring fight, gymnast performance
Nature (20 categories)
Waterfall forest sunlight, mountain snow peak, beach sunset ocean waves, desert sand dunes, volcano eruption, aurora borealis night sky, tornado storm, lightning thunderstorm, rainbow after rain, foggy forest morning, cherry blossom spring japan, autumn leaves forest, cactus desert, coral reef underwater, glacier ice arctic, canyon rock formation, lake reflection mountains, river rapids, rice fields asia, lavender field purple
Urban & Architecture (16 categories)
City skyline night lights, busy market street, bridge over river, skyscrapers downtown, old town cobblestone street, modern architecture building, church cathedral interior, mosque architecture, tokyo street night, paris street cafe, new york times square, london bridge, construction site workers, abandoned building, graffiti wall art, shopping mall interior
Food & Drinks (16 categories)
Pizza fresh oven, colorful fruit market, coffee latte art cafe, sushi restaurant, burger fast food, pasta italian restaurant, ice cream colorful, chocolate dessert cake, street food vendor asia, wine vineyard, bread bakery fresh, ramen bowl japanese, tacos mexican food, indian curry spices, salad healthy food, breakfast pancakes
Technology (12 categories)
Laptop coffee desk workspace, smartphone hand, server room data center, robot futuristic, electric car charging, solar panels roof, wind turbines field, factory automation, 3d printer working, drone flying camera, vr headset gaming, circuit board electronics
Space & Science (8 categories)
Galaxy milky way stars, planet earth space, rocket launch nasa, telescope observatory, laboratory scientist, microscope research, satellite orbit earth, moon surface
Art & Culture (10 categories)
Museum art gallery, concert crowd music, theater performance stage, film camera movie set, fashion runway model, wedding ceremony, festival celebration fireworks, carnival parade, street performer crowd, yoga meditation sunrise
## Dataset Structure

dataset/
├── images/
│   ├── dog_running_in_park/
│   ├── cat_sleeping_on_sofa/
│   ├── sports_car_highway_night/
│   ├── mountain_snow_peak/
│   ├── pizza_fresh_oven/
│   └── ...
│
├── metadata.csv
└── dataset-card.md

Each folder represents a **semantic category label**, and contains images related to that concept.

Example:

images/dog_running_in_park/
images/cat_sleeping_on_sofa/
images/pizza_fresh_oven/

---

## Metadata File

The dataset includes a **metadata.csv** file describing each image.

Format:

filename,label,filepath

Example:

dog1.jpg,dog_running_in_park,images/dog_running_in_park/dog1.jpg
pizza1.jpg,pizza_fresh_oven,images/pizza_fresh_oven/pizza1.jpg

This file allows the dataset to be easily loaded in machine learning pipelines.

---

## Data Collection

Images were collected using the **icrawler** Python library from **Bing Image Search**.

Each category corresponds to a search query used during crawling.
Example queries include:

* dog running in park
* sports car highway night
* waterfall forest sunlight
* pizza fresh oven
* galaxy milky way stars
* city skyline night lights

Images were automatically downloaded and organized into folders according to the query used.

---

## Intended Use

This dataset can be used for experiments involving:

* image retrieval systems
* CLIP embeddings
* multimodal search
* vector similarity search with FAISS
* computer vision experimentation
* machine learning research

---

## Limitations

* Labels are automatically generated from search queries and may contain noise.
* Image quality and resolution may vary.
* Some duplicate or near-duplicate images may exist due to automated crawling.

---

## License

Dataset structure and metadata are licensed under **CC BY 4.0**.

Images were collected from publicly available web sources using automated tools.
Original copyrights for individual images remain with their respective owners.

---

## Author

Created by **Darsh Vithlani (2026)**.